A GPU-based hydrodynamic numerical model for urban rainstorm inundation simulations

Abstract

The response capacities of urban flood forecasting and risk control can be improved by strengthening the computational abilities of urban flood numerical models. In this work, a GPU-based hydrodynamic model is developed to simulate urban rainstorm inundations. By simulating rainstorm floods in a certain area of Xixian New City, the established model can implement high-resolution urban rainstorm inundation simulations with significantly accelerated computing performances. The accelerated computation efficiencies of the different rainstorm event simulations under resolutions of 5 and 2 m are quantitatively analysed, showing that the absolute and relative speedup ratios for all scenarios of applying two GPUs range from 10.8 to 12.6 and 1.32 to 1.68 times as much as those of a CPU and a single GPU, respectively. The application of a large-scale rainstorm inundation simulation shows the excellent acceleration performance of the model compared to previous research. In addition, the greater the number of computational grids included in the simulation, the more significant the effect on the acceleration computing performance. The proposed model efficiently predicts the spatial variation in the inundation water depth. The simulation results provide guidance for urban rainstorm inundation management, and they improve the time and efficiency of urban flood emergency decision-making. HIGHLIGHTS The GPU-based hydrodynamic model is used to evaluate sponge city rainstorm inundation.; Significant speedup of the model executed on multi-GPU compared to a single GPU.; The model is efficient for performing high-resolution sponge city inundation simulation.; Validation and speedup computing performance tests on real urban city-scale areas.;

Keywords

• accelerated computing efficiency
• hydrodynamic model
• multi-gpu parallel computation method
• urban rainstorm inundation simulation